© 2005 American Thoracic Society doi: 10.1164/rccm.2506002
Inhaled Corticosteroids and Chronic Obstructive Pulmonary DiseaseNew and Improved Evidence?Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland As the epidemic of chronic obstructive pulmonary disease (COPD) has worsened over recent decades, therapeutic approaches have been developed for a disease that was previously considered to have an inexorable and irreversible course. The recognition of the extensive inflammatory process in the small airways, in addition to emphysema, has motivated studies of inhaled corticosteroids in patients with COPD (1, 2). Systematic reviews indicate that inhaled corticosteroids reduce the frequency of exacerbations (3) without altering the rate of decline of the FEV1 (4). For the key clinical outcomes of mortality and hospitalization, the evidence, coming largely from observational studies, has been mixed. This pair of endpoints is particularly critical to the evaluation of inhaled corticosteroids because a beneficial effect would indicate an improved natural history, the most critical objective of therapy. The mixed findings of the studies reported to date have many potential explanations, including variability among the populations studied and methodologic differences in the details of study design and analysis. In a 2003 publication in the AJRCCM, Suissa (5) identified one source of bias affecting some observational studies: a definition of use of inhaled corticosteroids that led to some of the follow-up time being event-free by definition, so-called immortal follow-up time. In an accompanying editorial, I commented on this bias and other methodogic issues in observational studies of therapeutic agents, including the need to contend with "confounding by indication" using appropriate adjustment methods, such as the propensity score (6). In this issue of the AJRCCM (pp. 460–464), Kiri and coworkers (7) report the findings of two distinct approaches to assessing the effectiveness of inhaled corticosteroids for reducing hospitalization and mortality, using the United Kingdom's General Practice Research Database (GPRD), a well characterized resource for pharmacoepidemiology (8, 9). In one approach, they created a matched cohort comprising 393 pairs, one treated and one untreated. Follow-up began on the same day for both, so as to avoid any immortal follow-up, and potential confounding characteristics were taken into account using the propensity score approach. In the second approach, a case-control study was nested within the GPRD cohort; patients had either been rehospitalized or had died, whereas the matched control subjects had not experienced either event. The association of inhaled corticosteroid use in the 6 months before the event was evaluated. With both approaches, inhaled corticosteroid use was associated with similar 30% reduction in the risk for outcome. By design, both approaches are free of any bias from immortal follow-up time, although the possibility of persistent confounding by indication, even with an attempt to control through adjustment, cannot be fully set aside. In the cohort study, the propensity score approach created matched groups that were more comparable than before matching (see Table 1 in Reference 7). However, the two groups were not comparable in all elements of the propensity score after the matching. For example, more patients in the steroid-treated group had a history of prior asthma and fewer had been on home oxygen. A more favorable profile to therapeutic response in the treated group could conceivably contribute to the estimated beneficial effect of inhaled corticosteroids. Since Suissa's 2003 article, two other analyses have been reported that addressed the immortal follow-up time issue. Suissa explored bias from immortal follow-up time in the Saskatchewan database in a cohort of 3,524 patients with COPD (10). His analytic approach was intended to examine potential bias from the design of a previous study based on the GPRD database (11). Mirroring the approach used for the earlier analysis of the GPRD database and then making accommodations for immortal follow-up time, an uncorrected, statistically significant benefit became much more modest and not statistically significant. In the second report, Fan and colleagues (12) performed a prospective cohort study of hospitalization and death in 8,033 veterans, one-third of whom had received inhaled corticosteroids. Their analytic approach used a time-variable classification of medication use and adjusted for confounding using a set of appropriate clinical and demographic characteristics. The time-dependent analyses did not find a statistically significant benefit of inhaled corticosteroids, although the estimated hazard ratios were below 1 (i.e., indicating a reduced risk). In 2003, I commented that the weight of evidence from observational and experimental studies demonstrated "little indication at present of benefit" for use of inhaled corticosteroids and the outcomes of hospitalization and mortality (13). The newer studies provide further mixed evidence, although their methods are dealing with immortal follow-up time and bias from confounding by indication, to the extent possible. Reflecting the still incomplete and evolving evidence base, some recent recommendations on use of inhaled corticosteroids in COPD have been guarded (14, 15). Further evidence is needed, both from clinical trials with sufficient sample sizes to address hospitalization and mortality and from rigorous analysis of administrative databases. The TORCH (Toward a Revolution in COPD Health) study, with 6,000 participants, may provide informative results. Another approach for gaining information is pooling of data from individual trials; at the 2005 American Thoracic Society meeting, a reduction of all-cause mortality associated with inhaled corticosteroid therapy was reported based on 5,082 participants in seven trials (16). Pooled analyses of the observational databases should also be performed to explore the basis of the heterogeneous findings to date and to assess findings with harmonized methodologies. The still-evolving evidence does provide some indication of the potential magnitude of benefit associated with inhaled corticosteroids for treatment of COPD. The new report by Kiri and coworkers (7) suggests a 30% reduction in risk for rehospitalization or death, a meaningful gain with COPD. Small studies lacking statistical power could readily be interpreted as null for an effect of this magnitude. More precise estimates of any benefit will only be gained through "large N studies" and evidence synthesis. Careful attention to methodologic details is also needed; the work of Suissa and the new report by Kiri and colleagues show that the methodologic challenges of studying COPD outcomes can be met. FOOTNOTES Conflict of Interest Statement: J.M.S. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. GlaxoSmithKline provides unrestricted support for tobacco control to the Institute for Global Tobacco Control Center at the Bloomberg School of Public Health, which is codirected by J.M.S. REFERENCES
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
